# Decision models, part two: train your own (slim) > Vlad Cealicu on the other way out of the decision model land grab. Coinbase fine-tuned Qwen3.5-9B on its own fraud data and says it beat frontier models. Ten of the fourteen decision models we list publish weights, and why a company's own decisions are the best training data it has. - Full: https://www.anchorterminal.com/blog/decision-models-train-your-own.md (~1,700 tokens) · this version ~830 tokens · JSON https://www.anchorterminal.com/blog/decision-models-train-your-own.json · canonical https://www.anchorterminal.com/blog/decision-models-train-your-own - Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · Updated: 2026-10-12 By Vlad Cealicu, 12 October 2026. 726 words, about 3 min to read. > This is part two. [Part one, Decision models are popping up like mushrooms after the rain](/blog/decision-models-everyone-wants-your-decisions), is the problem. Fourteen decision models in four weeks, and every hosted one gets to see the decisions your agents make. This one is the solution. ## What Coinbase did [Part one](/blog/decision-models-everyone-wants-your-decisions) ended on the problem, that every hosted decision model sees what you were choosing between, and a few days later Brian Armstrong posted the solution I hadn't written yet. He thinks it's a pattern. He didn't publish the numbers, so I can't check them (and fraud numbers are the last thing a company publishes). ## Same base model, different owner Microsoft-Decision-1, which launched last Friday, is post-trained from the same Qwen3.5-9B [[2]](#ref-2). That's part one's argument in a single base model. ## Most decision models are already open - 10: publish weights you can download - 7: are built on a Qwen base - 4: are hosted only - $1: for a Kev-4B training run on an H100, by its README The 14 decision models we list, 12 October 2026 The four hosted only are OpenAI's Decisions API, Jev, Celeris-1 and Microsoft-Decision-1 [[3]](#ref-3). Clef and Kev also answer the same request as Jev's hosted API, so a team can start on a hosted model this week and move to its own fine-tune later by changing a base URL [[7]](#ref-7). ## Why decisions are the easy thing to fine-tune A decision is a narrow task with a fixed set of answers (fraud or not, approve or escalate, which of five queues). It's also the work most companies already pay people and rule engines to do. ## The catch A fine-tuned model is a thing you now own, which means you evaluate it, watch it and retrain it when the data moves. So I think most companies end up with both. ## Where to look The [decision model shortlist](/best/decision-models/) and [every head-to-head](/compare/decision-models/) have all fourteen, and each listing says whether its weights are open, under what licence and on which base model. --- - Older: [The Log, issue 1: decide, declare, depart](https://www.anchorterminal.com/blog/the-log/issue-1-decide-declare-depart.md) - All posts: https://www.anchorterminal.com/blog/index.md ## References Read on 12 October 2026. Quotes are as published. [1] Brian Armstrong on X, October 2026. Source for Coinbase's fine-tune of Qwen3.5-9B for fraud prevention, the quotes and the internal command. [2] Anchor Terminal, Microsoft-Decision-1. Source for its Qwen3.5-9B base and its launch, with Microsoft's pages. [3] Anchor Terminal, decision models, 12 October 2026. Source for the fourteen models, which publish weights, their licences and their base models. [4] Kev, README on GitHub. [5] Anchor Terminal, Laya. [6] Anchor Terminal, Strands Decider 2B. [7] Anchor Terminal, Jev.